EU AI Act: Claude's Text Watermarking Begins
Anthropic's AI model will generate text with a hidden watermark to comply with EU regulations, using the SynthID-Text approach to modify low-stakes word choices

Anthropic's AI model Claude will start generating text with a watermark to comply with the EU AI Act, which requires AI companies to use systems that make it possible to identify AI-generated content. The watermark is created by modifying low-stakes word choices in a way that is undetectable to readers but can be detected with a digital key.
The watermark does not impact the quality of Claude's output, and the difference between watermarked and un-watermarked text will not be distinguishable to readers. Anthropic will use the SynthID-Text approach for watermarking, which involves influencing inconsequential word choices made by its models.
The EU AI Act requires AI companies to use systems that make it possible to identify AI-generated content, and Anthropic is implementing this change to comply with the regulation. Other major model developers have also signed the same Code of Practice and will be implementing their own watermarks.
Anthropic plans to release a watermark detection API, which will allow users to detect the watermark in text generated by Claude. However, light editing of watermarked text may not completely remove the watermark, while a complete rewrite of watermarked text will likely remove the watermark.
## What is watermarking? Watermarking is a technique used to apply an identifier to electronic data. In the case of Claude, it involves modifying low-stakes word choices in a way that is undetectable to readers but can be detected with a digital key. This is done by influencing the model's predictions of the next word in a sequence of words.
## Why it matters The EU AI Act's requirement for AI companies to use systems that make it possible to identify AI-generated content is a significant development in the regulation of AI. The use of watermarks like the one implemented by Anthropic will make it possible to detect AI-generated content and ensure transparency in the use of AI. This will have implications for a wide range of industries, from media and entertainment to education and research.




